Two people measure the same fabricated part and get two different answers. The part has not changed. The measuring equipment is fine. What differs is where each of them decided the part begins, and that decision is a datum. A datum is not a line on a drawing for its own sake. It is a statement about which surface the part performs its job against.
This is why datum arguments are rarely about measurement technique. They are about intent. If a bracket bolts to a machined face and locates against a dowel, then that face and that dowel define the part's position in the world, and every dimension that matters should be referenced from them. A datum chosen because it was convenient to draw, rather than because it is where the part touches its neighbour, produces a drawing that can be inspected perfectly and still not fit.
The useful habit is to ask, for every part, which surfaces are the ones that decide where it sits. Usually there are only a few, and they are almost never the largest or the most attractive face. They are the ones in contact.
The convenient datum is the one that makes the dimensioning easy to draw. It might be a nice straight edge, a centreline, or the origin of the model. The functional datum is the one that makes the part work. On simple parts the two coincide, which is why the distinction gets overlooked until a part arrives that separates them.
An exhaust flange is a clear case. The gasket face decides where the part seals, and the port spacing decides whether it bolts up. A drawing that dimensions from the outer profile of the flange, because that is the easiest thing to pick up in a model, is dimensioning from a surface that does nothing. The flange profile is a trimming operation. The face and the port pattern are the function.
If the part is located on a non-functional surface for inspection, the measurement reports the variation between two surfaces rather than the variation that matters. The part can pass inspection and fail to fit, or fail inspection and fit perfectly. Both outcomes erode trust in the numbers, and the usual response is to tighten tolerances, which makes the problem more expensive without making it more correct.
The same failure appears in fabrication. A weldment tacked up in a fixture that locates on an arbitrary face will distort relative to that face rather than relative to the mounting points. The fixture is doing exactly what it was told. It was simply told the wrong thing.
A welded assembly is harder to datum than a machined part because the surfaces it should be located on may not exist until after welding, and welding moves things. The practical answer is to locate on the features that will be machined or fastened afterwards, and to accept that the fabricated surfaces are intermediate. This is why so much welded work carries a note about machining after welding rather than before.
Where no machining follows, the datum usually has to be a pair of features that are welded in the same operation and therefore move together. Choosing two surfaces on opposite sides of a seam that is welded last is asking for a datum that shifts depending on which tack held.
It also helps to say which surfaces are allowed to be non-critical. A drawing that datums everything datums nothing, because there is no guidance about which face wins when two of them disagree.
A part located on more surfaces than its function requires is over-constrained. In practice that means the part cannot sit flat on all of them at once, because real surfaces are not perfect. Something has to give, and what gives is decided by the clamping order rather than by the drawing. Two inspectors clamping in different sequences will report different results from the same part.
The classic symptom is a part that rocks. It touches on three points and lifts at the fourth, and the measurement depends on which corner was tightened first. That is not a manufacturing defect. It is a statement that the part has been asked to conform to more geometry than any real object can.
When a feature is located with a clearance hole, the part can move within that clearance before the fastener is tightened. That movement is datum shift, and it is a real allowance that has to be shared between the parts being joined. A drawing that assumes zero shift while specifying clearance holes is describing two incompatible things at once.
The consequence is that hole patterns which look generous on paper fight during assembly. The fix is rarely a tighter hole. It is usually to decide which part is the master, locate the mating part from the master's features, and accept that the clearance exists so the assembly can find its own position. Understanding that shift is what makes a fixture design sensible instead of merely precise.
A datum has to be measurable with the equipment in the building. A theoretical centreline is a fine reference for dimensioning, but if the only way to establish it is a machine nobody has, then the dimensions built on it cannot be verified. Datums that are real surfaces, or real features that can be indicated, get used. Datums that exist only in a model get ignored, and ignored dimensions are not requirements.
There is also a physical question about stability. A datum surface that is thin, unsupported, or adjacent to a weld will move. Choosing it means the measurement changes depending on when it is taken, which is the fastest way to make two people disagree about a part that is actually identical.
The first article is where datum choices stop being theoretical, so it is worth confirming these points before anything is indicated.
Written down, that list is short. It is also the difference between a first article that settles an argument and one that starts it.
Given the same part, two experienced people will still disagree if one measures from a machined face and the other from a trimmed edge. Neither is careless. They are answering different questions, and the drawing only asked one of them. The disagreement is evidence that the drawing left the datum implicit and each reader supplied their own.
That is worth remembering when a supplier reports a dimension as out. The useful first question is not how far out it is, but where it was measured from. A dimension that is out when taken from the functional datum is a real problem. The same dimension taken from a convenient edge is a conversation about references.
Our own work is organised around the material and the joint rather than the drawing frame, which is why the datum logic tends to be visible in the parts themselves. You can see it across the titanium fabrication components and TIG titanium welding parts ranges, and in the way the stainless steel exhaust system assemblies are built around their mounting faces. The exhaust pipe work shows the same idea in tube, where the centreline is functional and the outside surface is not.
How a first article gets agreed is set out under customization, and the material behaviour that decides how much a datum surface will move sits under shop by material. A datum is cheap to choose and expensive to change, which is exactly why it deserves the argument early.
